Solving the redundant optimization problem for human muscles depends on the cost function. Choosing the appropriate cost function helps to address a specific problem. Muscle synergies are currently limited to those obtained by electromyography. Furthermore, debate continues regarding whether muscle synergy is derived or real. This study proposes new cost functions based on the muscle synergy hypothesis for solving the optimal muscle force output problem through musculoskeletal modeling.We propose two new computational cost functions involving muscle synergies, which are extracted from muscle activations predicted by musculoskeletal modelling rather than electromyography. In this study, we constructed a musculoskeletal model for simulation using the "Grand Challenge Competition to Predict In Vivo Knee Loads" dataset. Muscle synergies were obtained using non-negative matrix factorization. Two cost functions with muscle synergies were constructed by integrating the polynomial and min/max criterion. Two new functions were verified and validated in normal, smooth, and bouncy gaits.The muscle synergies based on normal gaits were classified into four modules. The cosine similarities of the first three modules were all >0.9. In the normal and smooth gaits, the forces in most muscles predicted using the two new functions were within three standard deviations of the root mean square error for electromyographic comparisons. Predicted muscle force curves using the four methods as well as characteristic points (i.e., time points in the gait cycle when the significant difference was observed between normal and bouncy gaits) were obtained to validate their predictive capabilities.This study constructed two new cost functions involving muscle synergies, verified and validated the ability, and explored the potential of muscle synergy hypothesis.